Sunday, March 10, 2013

Causal follies: gun control edition

The Atlantic Cities column summarizes a study recently published in (ahem) JAMA Internal Medicine.

The study examines cross-sectional correlations between state gun control law strictness and firearms deaths. I have some concerns:

1. An alternative model would be that gun control law severity is a function of the number of firearms deaths. That is, perhaps the causality runs the other way. Or both ways. The cross-sectional "selection on observed variables" design can shed little light on this though some actual leg work on the legislative histories could..

2. The model employed in the study conditions on firearms ownership and on "other violence-related deaths". The first of these is an intermediate outcome, the second is jointly determined with the dependent variable to the extent that firearms and other tools for offing people are substitutes. Nether should be on the right-hand side of the model.

The Atlantic writer is very coy about causality and quotes the author as cautioning that the estimated relationship is only correlational. But that does not stop the Atlantic writer from making a strong causal claim in the last line of the article. I give the Atlantic writer an F.

Finally, I think that my concerns above make it clear why this was published in a medical journal, which is likely ill-equipped to evaluate a study such as this rather than, say, a clinical trial, rather than in a social science journal. This study also reinforces the point I made in an earlier post about it being a bad idea to treat guns as a public health issue rather than a social science one.

No comments: